• Aucun résultat trouvé

Indecomposable projective modules on affine domains

N/A
N/A
Protected

Academic year: 2022

Partager "Indecomposable projective modules on affine domains"

Copied!
19
0
0

Texte intégral

(1)

C OMPOSITIO M ATHEMATICA

V. S RINIVAS

Indecomposable projective modules on affine domains

Compositio Mathematica, tome 60, n

o

1 (1986), p. 115-132

<http://www.numdam.org/item?id=CM_1986__60_1_115_0>

© Foundation Compositio Mathematica, 1986, tous droits réservés.

L’accès aux archives de la revue « Compositio Mathematica » (http:

//http://www.compositio.nl/) implique l’accord avec les conditions gé- nérales d’utilisation (http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une in- fraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.

Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques

http://www.numdam.org/

(2)

115

INDECOMPOSABLE PROJECTIVE MODULES ON AFFINE DOMAINS

V. Srinivas

e Martinus Nijhoff Publishers, Dordrecht - Printed in the Netherlands

Let A be an affine domain i.e. an

integral

domain of finite

type

over an

algebraically

closed field k. If A has Krull

dimension d,

then

by

the

stability

theorem of Bass

[Ba],

any

indecomposable projective

A-module

has rank at most d. In this paper, we consider the conditions under which A has

indecomposable projective

A-modules of rank d. Our main result is:

THEOREM: Let k be a universal

domain,

and assume that d = dim A 3.

Assume that

Spec

A has at worst isolated normal

singularities; further if

dim

A = 3,

then

char,

k =1=

2, 3,

5. Then there exist

indecomposable projective

A-modules

of

rank d

if

and

only if F0K0(A) ~

0.

Here

F0K0(A)

is the

subgroup

of the Grothendieck group of

projec-

tive A-modules

generated by

the classes of residue fields of smooth

points;

it is

isomorphic (upto 2-torsion)

to the Chow group of zero

cycles

of

Spec

A

(see §1

for the

definition).

The restriction to

char,

k =1=

2, 3,

5 if

d = 3 is because we need to use the theorem of resolution of

singularities

for

3-folds,

which is

proved

under the above restriction in

[A].

In a

footnote in their paper

[MK],

M.P.

Murthy

and Mohan Kumar state that

Abhyankar

can now prove resolution for 3-folds in all

characteristics;

assuming

this we may

drop

the

hypothesis

on the characteristic.

The

hypothesis

on the

singularities

is needed for technical reasons in the

proof,

but may be unnecessary: 1 do not know any counter

example

for non-normal A or for normal A of dimension 3 with non-isolated

singularities. However,

it seems

plausible

that k must be a

’big’

field in

order that

F0K0(A)

be

non-zero; I

do not know an

example

of an affine

domain of

dimension

2 over Qu with

F0K0(A) ~

0.

Many examples

exist of such affine domains over the

algebraic

closure

of Q(x, y ) (x, y indeterminates)

with

F0K0(A) ~

0. But our methods do not seem to be

directly applicable

to such

rings

*. We discuss

examples

of

rings

with

F0K0(A) ~

0 in

§4

and

§5.

The

proof

is based on the

following

idea.

Suppose

P is

projective

of

rank d = dim

A 2,

and for a suitable notion of the Chow

ring together

* One can combine our result with the method of [BS] to get such results.

(3)

with a

theory

of Chern

classes,

we have

C1(P) =

...

Cd-1(P)

= 0. Then

if

P = Q

~ L where L E Pic

A, rank Q

= d -

1,

we have an

identity

in

the Chow

ring

This forces

Ci(Q) = (-1)iC1(L)i,

and

Cd(P) = (-1)d-1C1(L)d.

Thus

our Theorem will follow once we show that if

F0K0(A) ~ 0,

then there exists a

projective

module P with

Ci(P) = 0

for

i d,

and such that

Cd(P) ~

0 is not

rationally equivalent

to a

cycle

of the form

(-1)d-1C1(L)d.

The idea of the

proof

is to show that if

F0K0(A) ~ 0,

then the Chow group of

0-cycles CHd(A)

is ’infinite dimensional’ in a

suitable sense, while the classes of

cycles

of the form

(-1)d-1C1(L)d

lie

in a countable union of

’algebraic

sets’ in

CHd(A),

each

component

of which has dimension bounded

by

a number

depending only

A

(this

relies

on the

theory

of the Picard

variety).

This is done

using

ideas of Roitman

[R]

which have to modified somewhat to deal with

affine, possibly singular

varieties

(Roitman basically

deals with smooth

projective

varie-

ties).

Roitman’s ideas

directly apply

when A is smooth of dimension

2,

but our modifications seem to be needed even for smooth A if dim A = 3.

Finally,

we do not know if the theorem

ought

to be true if dim

A = 4, though

we know no counter

examples.

If we

try

an

analogous argument,

and let P be

projective

of rank 4 with

Cl ( P )

=

C2 ( P )

=

C3 ( P )

=

0,

then

suppose that P

= P1 ~ P2

with

rank Pi

= 2. A Chern class

computation yields C1(P2) = - C1(P1), C2(P2)

=

C1(P1)2 - C2(P1),

and

C4(P)

= 1 2C1(P1)4 - C2(P1)2.

The

0-cycles

of the form

C1(P1)4

range over a countable union of

algebraic

varieties of bounded dimension.

However,

CH2(A)

could be infinite

dimensional,

so we seem to have no control on

the set of classes of the form

C2(P1)2 in CH4(A).

This

problem

was

suggested

to the author

by

M.P.

Murthy,

in the case

d = 2. I wish to thank N. Mohan Kumar and M.V. Nori for

stimulating discussions;

in

particular,

Mohan Kumar

suggested

the trick of

looking

at

projectives

with

Ci

= 0 for i

d,

which reduces the

problem

to line

bundles even if d = 3.

§1. Equivalence

relations on zéro

cycles (af ter Roitman)

Let X be a normal

projective variety

over an

algebraically

closed field

k,

and let U c X be the set of smooth

points.

The group of zero

cycles

on

X,

denoted

by Z0(X),

is defined to be the free abelian group on the

points

of U. Let

R(X)

c

Z0(X)

be the

subgroup

of

Zo ( X) generated by cycles

of the form

(f ) c

where C c X is a curve with

C ~ (X - U) = Ø

and

f

is a non-zero rational function on

C; (f)C

denotes the divisor of

f

on C

(note

that

C ~ ( X - U ) = Ø just

means C c

U;

we write it as above

to

emphasise

that C is a closed subset of

X).

The Chow group of

(4)

0-cycles

is defined to be the

quotient Z0(X)/R(X);

we denote it below

by A0(X) (rather

than

CHd(X),

where d = dim

X). Similarly

if A is a

normal affine

domain,

let V c

Spec

A be the set of smooth

points, Zo ( A )

the free abelian group on

points

of

V,

and

A0(A)

the

quotient

of

Z0(A)

modulo the group

generated by (f)C

where

C c Spec A

is a curve,

C ~

((Spec A) - V) = Ø,

and

f

is a non-zero rational function on C.

If

X,

U are as

above,

the n th

symmetric product sn(u) parametrizes

effective zero

cycles

of

degree n

on X i.e.

cycles 03A3ni(Pl)

with

ni &#x3E; 0,

03A3ni

= n,

P, e

U. There are natural maps

where

[ A denotes

the zero

cycle corresponding

to A E

Sm(U).

In

[S1],

Lemma

(1.1)

we showed that if

is induced

by ’¥ m n’

then

03B3-1m,n(0)

is a countable union of

locally

closed

sets in

Sm(U) Sn(U),

and used this to

study

’infinité

dimensionality’

of

A0(X).

Here we

study

infinite

dimensionality

for

cycle

groups associ- ated to more

general equivalence

relations.

Let

W ~ Z0(X)

be a

subgroup.

Then W defines an

equivalence

relation on

0-cycles, namely [A] ~ [ B iff [A] - [ B ] ~

W. Let

AW0(X) = Z0(X)/W

be the

corresponding quotient.

We define W to be admissible if

(i)

W ~

R(X) (ii)

for any m, n if

crm,n: Sm(U)

X

Sn(U) ~ AW0(X)

is

induced

by ’¥ m,n’

then

~-1m,n(0)

is a countable union of

locally

closed

sets. *

We recall some

terminology

of Roitman. For a

variety X

over a

universal

domain,

a c-closed set in X is a countable union of closed subsets of X. For any c-closed set F =

U Fi

in

X,

we define dim F = sup

(dim Fi).

This

concept

is well

defined,

and

depends only

on the set F

and not on the

specific decomposition

into sets

Fi (since

an irreducible

variety

over a universal domain is not the union of a countable set of proper

subvarieties). Similarly,

a

c-locally

closed set in X is a countable

union of

locally

closed subsets of X. For any

c-locally

closed

set

G =

UGi

with

Gi locally closed,

we define the c-closure of G to be

UGil

This is

also characterised as the smallest c-closed subset of X

containing G,

since if F =

U Fi :)

G where

F

are

closed,

then each

Gi

is contained in a

finite

union of the

F,

which is a closed set

(and

so contains

Gi).

In

particular

the c-closure of G does not

depend

on the

decomposition

of G

* By the methods of [LW] one can prove that

~-1m,n(0)

is a countable union of closed sets if

W is admissible, but we do not need this.

(5)

into

locally

closed subsets. For any

c-locally

closed set

G,

we define dim G to be the dimension of its c-closure.

Finally,

a c-open set in X is defined to be the

complement

of a c-closed set.

We have the

following lemma, essentially

due to Roitman.

LEMMA

(1.1):

Let X be a

variety

over a universal

domain,

r c X X X the

graph of

an

equivalence

relation on X.

Suppose

that r is

c-locally

closed in

X X X. Then there is a c-open set V c X and an

integer d

0 such that

for

any x E

V,

the set

is a

c-locally

closed subset

of

X

of

dimension

d,

such that all irreducible components

of

the c-closure

of rx

have dimension d.

(We

note that every c-closed set in X has a

unique

irredundant decom-

position

into irreducible closed sets, so it makes sense to talk of the

components

of a c-closed

set).

PROOF: Let

r

be the c-closure of r. Then

r

is also the

graph

of an

equivalence

relation on X. One

sees easily

that there

exists a

c-closed set

F ~ X such that

for x ~ X - F, (0393)x = {y ~ X|(x, y) ~ 0393} is just

the

c-closure of

0393x.

Hence we reduce to the case when r itself is

c-closed,

and so

rx

is also c-closed. Let r =

U0393i

be an irredundant

decomposition

into irreducible closed subsets of X X X.

We claim first that there exists a

unique component

of r

containing

the

diagonal

in X X X.

Indeed,

there is a proper c-closed subset F c X such that for all x

E X - F,

either

(i) ({x} 0393x) ~ 0393l is empty, or

(ii)

if

0393i,x

=

{y ~ X|( x, y ) E ]ri 1,

then

fi x

is a finite

disjoint

union of

irreducible subvarieties of dimension

equal

to

(dim rl -

dim

X);

further

rx = U0393i,x is an

irredundant

union,

where i runs over

indices such that

]ri @ x :gÉ 0 (by irredundancy

we mean that no

component

of

rl,x

is contained in

0393J,x for j ~ i).

Now

suppose y E 0393i,x - U 0393J,x -

F. Then

( x, y ) E ]ri - U rj.

For

any y E rx,

we have

ry

=

lx

as r is an

equivalence

relation. If

r y = U0393i’,y

is the above

decomposition (which

exists

because y 5É F),

then the

component

of

0393l,x containing y

lies in a

unique 0393i’,y,

so that

( y, y) E fi,

-

U0393J.

Thus

rt,

is the

unique component

of r

containing

the

diago-

j 4= i,

nal.

Further, Fi,x

and

r,,

are

equidimensional

schemes with a common

irreducible

component (namely,

the

component through y )

and so

dim

rl = dim 0393i,x + dim

X = dim

0393i’,y

+ dim X = dim

0393’i.

(6)

Since i was

arbitrary,

dim

rl

is

independent of i,

at least for all

components fi

which dominate X under the first

projection.

Thus

dim

fi,x

= dim

0393J,x

for all such

fi, 0393j

i.e.

rx

is a union of irreducible varieties each of which has dimension d =

(dim Fi,

- dim

X).

This proves the lemma.

Now let W be an admissible

equivalence

relation on

0-cycles on X,

where X is a normal

projective variety

over a universal domain. Then via ~m,n, W induces a

c-locally

closed

equivalence

relation on

Sm(U)

X

sn(u),

where U = X -

Xsing.

Let

W m, n

be the number d

given by

Lemma

(1.1)

in this context, and let

We define

SD(U)

to be a

point, corresponding

to the zero

cycle 0,

and let

dn

=

dn,o,

wn =

Wn,o

etc.

PROPOSITION

(1.2) :

There are

integers dW, jW

and no &#x3E; 0

depending only

on X and W such that

0 dW(X)

dim

X, and for

m

+ n n0

we have

PROOF: We first show that

Wm,n = wm+n. Indeed,

consider

Clearly ~m,n((A, B)) = crm,n((C, D))

iff

~m+n(A

+

D)

=

~m+n(B

+

C)(~m+n

=

~m+n,0).

Now rm,n is the

graph

of the

equivalence

relation determined

by ~m,n,

so that if

0393m,n0

is the

unique component

of the c-closure of

0393m,n containing

the

diagonal,

then

Let 0393m+n ~

Sm+n(U)

X

Sm+n(U)

be the

graph

of the

equivalence

rela-

tion determined

by

~m+n, and let

0393m+n0

be the

unique component

of the c-closure

containing

the

diagonal;

then

Let NF:

Sm(U)

X

Sn(U)

X

Sm(U) Sn(U) ~ Sm+n(U)

X

sm+n(u)

be

given by (A, B, C, D) - ( A + D,

B +

C).

Then 03A8 is

finite,

and

03A8-1(0393m+n)

= 0393m,n. Hence dim

0393m,n0 =

dim

0393m+n0 (since 03A8

maps the

diagonal

to the

diagonal)

i. e.

Wn,n

= wm+n, and

dm,n = dm+n.

Let Vn ~ U

Sn(U)

be the set

(7)

Similarly

for A E

sn(u)

let

VAn

be

(the image

in U

of)

the fiber over A of

the

projection

Vn ~

Sn(U).

For any n, V" is a

c-locally

closed set in

U X

Sn(U),

and for any

A, VAn

is a

c-locally

closed set in U. Thus for

some

non-empty

c-open set

U1

~

Sn(U)

and for all A E

Ul,

we have

Let 03B4n

= dim

hA

for

any A E UI.

For any A E

Sn(U)

and any x E U we have

VnA ~ Vn+1A+x,

so that

03B4n 03B4n+1 dim X.

Thus

03B4n = 03B4n+1

for all

n n0. Let 03B4 be this common

limiting value,

and let

dW =

dim X - 03B4.

Consider the map

(addition)

If

0393n+1A = {B ~ Sn+1(U)|(A)=~n(B)}

for

any A ~ Sn+1(U),

then

p1(03C0-1(0393n+1A))

=

Vn+1A,

where p1 is the

projection

U X

Sn(U) ~

U. For

x E

Vn+1A,

the fiber of

(p1)| 03C0-1(0393n+1A) just 0393nB,

where B is the effective

0-cycle A-[x].

Since w n = dim

0393nB

for B in some c-open subset of

U,

and

similarly wn+1 = dim 0393n+1A

for

general A,

we have

wn+1 = wn + 03B4

i.e.

dn+1 = dn

+

dW for n

n o . This proves the

proposition.

REMARK: This

proof only

used the

property (ii)

in the definition of an

admissible

equivalence

relation.

LEMMA

(1.3):

Either

d w = 0 or dW 2. If dW = 0, then for

some n

image(~n,n) is a subgroup of AW0(X)

with

cyclic quotient.

PROOF:

Suppose dW

1. Then

for n

n0, we have dim

Vn+1A

dim X

- 1 for

any A E Sn+1(U), by

definition of

dX =

dim X - 8. Thus there

exists a smooth curve C c X with

C n (X - U) = Ø

and Cn

Vn+1A ~ Ø

for all

A ~ Sn+1(U) (in

fact any

general

smooth

complete

intersection

curve C will

do).

Then for any A E

Sm(U)

with m &#x3E; n0 we can find

Al

E

Sn0(U)

and

A2 E sm-no(C)

such that

Since W is an admissible

equivalence relation,

we have

diagrams

for all n

If we choose a base

point Po

E

C,

and g = genus

(C),

then

by

the

(8)

Riemann-Roch

theorem,

for n &#x3E; g and

any B ~ Sn(C),

there exists

B 1 E S g( C )

such that

Thus for any m &#x3E; n o + g, and

any A E Sm(U),

there exists B ~

Sn0+g(U)

such that

Thus

image ~m,m = image ~n0+g,n0+g

f or

m &#x3E; n0 + g

Le.

dX = 0

and

image

~m,m is a

subgroup

of

AW0(X)

whose

quotient

is the

cyclic

group

generated by

the

image of ~1(P0).

We recall another définition. For any

variety Z,

a set theoretic map

Z ~ AW0(X)

is defined to be a

regular

map if there exists a

diagram

where

f,

g are

morphisms

and

f

is

surjective.

Since W is

admissible,

if

C c X is a smooth curve with

C ~ (X - U) = Ø,

the natural homomor-

phism

Pic C -

AW0(X)

is a

regular

map. Given any finite set of

points

in

U,

there is a smooth curve C as above

through

these

points.

Thus any

given

zero

cycle

is in the

image

of such a

regular homomorphism.

Since

the jacobian J(C)

is

divisible,

if

AW0(X)’ ~ AW0(X)

is the

subgroup generated by cycles

of

degree 0,

then

AW0(X)’

is divisible.

Clearly ~n,n

factors

through AW0(X)’.

PROPOSITION

(1.4):

Let

dW = 0.

Then there is a

surjective regular

homo-

morphism f : A ~ AW0(X)’,

where A is an abelian

variety

which is a

quotient of Alb(X), and f

has countable kernel.

PROOF: We first remark that there is a

surjective regular homomorphism f-: 4 ~ AW0(X)

from an abelian

variety A,

such that ker

f

is countable.

This is

proved

for

W = R(X) (i.e.

rational

equivalence)

in

[SI],

Lemma

(1.3).

But the

only property

of

R(X)

used in the

proof

is that it is admissible in our sense.

Given this

homomorphism,

choose a base

point P E U,

and let r c U X A be the set

Then r is

c-locally closed,

and since

f

is

surjective,

r

surjects

onto U

(9)

under the

projection.

Hence some component

ro

of the c-closure of r

dominates U. Let

rl

=

ro

n

r;

then

03931 ~

U has

countable,

and hence

finite,

fibers

(since f

has countable

fibers).

There is an open set

Uo

c U

such that if

r2

=

rl X U Uo,

then 03C0 :

r2

~

Uo

is a finite flat

morphism

of

varieties. Let n be the

separable degree

and

p m

the

inseparable degree

of

this

morphism.

If x E

Lo

is

general,

then

03C0-1(x) = {(x, a1),..., (x, an)}

f or n distinct

points

a1,...,an E A . Then

x ~ pm(a1

+ ···

+ an) gives

a

well defined

morphism 03A8 : U0 ~ A

such that

f 0 ’l’: U0 ~ AW0(X)’

is the

natural map

crl,l multiplied by n. pm.

Hence

f o 4,(U.) generates AW0(X)’

(since AW0(X)’

is

divisible,

and

~1.1(U0 {P}) generates AW0(X)’).

Since

f

has countable

fibers, 03A8(U0) generates

A.

Regarding ’1’

as a rational

map

X ~ A, 03A8

factors

through Alb(X) by

the universal

property

of the Albanese

variety.

Since the

image

of

Alb(X) ~ A

contains

41(U), Alb(X) ~ A

is

surjective.

We

apply

these results to the

following

situation. Let X be a normal

projective variety

over a universal

domain,

and let V ~ X be an open subset

containing Xsing.

Then we have an exact sequence

where

A0(V)

is the free abelian group on

points

of

V - Vsing

modulo the

subgroup generated by cycles (f ) c

where C ~ V is a

(closed)

curve, C ~

Vsing = Ø,

and

f

is a non-zero rational function on C. The above

presentation

for

Ao(h)

follows

because

=

Xsing.

From the

presenta-

tion,

if we set

AW0(X) = A0(V),

then W is an admissible

equivalence

relation.

LEMMA

(1.5): Suppose

dim X

2, dW

= 0 and X - V generates Alb X.

Then AW0(X) = 0.

PROOF; Clearly AW0(X) = AW0(X)’,

so it suffices to show that if

f : A -

AW0(X)’

is a

surjective regular homomorphism

with countable

kernel,

then A = 0. As in the

proof

of

Prop. (1.4),

we claim there is a

diagram

A

priori

we

only

have such a

diagram

with

Uo

instead of

U,

but the rational

map X -

Alb X with P ~ 0 is a

morphism

restricted to U. Thus

we would have two

regular

maps

U ~ AW0(X)’, namely n.p.m ~1,1

and

f -

h ° g, which agree on

Uo.

We claim that

they

must agree

everywhere

on U.

Indeed,

if C c U is a

complete

smooth curve with

C ~ U0 ~ Ø,

then

(10)

we have 2

regular

maps C -

AW0(X)’ namely npm~1,1 | C

and

f

° h °

g|C.

Both induce

regular homomorphisms

Pic

C ~ AW0(X)’,

the former be-

cause W is

admissible,

and the latter because

Pic°C

= Alb

C,

so that

g c:

C ~ A induces a

homomorphism

Pic C ~ A. These two maps Pic C

~

AW0(X)’

are

equal,

because Pic C is

generated by

the classes of

points

in C ~

Uo.

Thus the two

original

maps on U agree on C. But such a curve

C can be found

through

any

point

of

U,

since X is normal. *

Now X -

Vc U,

and

f

o h o

g(X - V)

= 0 in

AW0(X) (by

definition

of

W,

and the

equality

of the 2

maps).

Thus h o

g(X - V) ~ (ker f)

which is a countable

subgroup

of A i.e. h o

g( X - V)

is finite. But

g(X - V)

c Alb X

generates Alb X, by

choice

of V,

and Alb X ~ A.

Hence A = 0 i.e.

AW0(X)-

= 0. This proves the lemma.

§2.

Families of

0-cydes

associated to Picard families

Let X be a normal

projective variety

of dimension d over an

algebrai- cally

closed field

k,

and let T be a normal

variety.

Assume

given

L~

Pic(X T),

and for each t ~

T,

consider the rational

equivalence

class of the zero

cycle (-1)d-1Ci(Lt)d ~ A0(X),

where

Lt = L|X {t}.

Here if X is

smooth,

the intersection

theory

is the usual one

(see

Fulton

[F])

while if X is

singular,

we use the

theory

of Levine

[L1] (we

discuss

this

theory briefly

in

§3).

In this

section,

the

only property

of the intersection

theory

needed is that if

D1,..., Dd

are effective Cartier divisors on X

meeting properly,

and

r1 supp Dl

c

U,

then

D1

~ ...

~Dd (as schemes) represents

the rational

equivalence

class of

[D1][D2] ... [Dd]

in

A0(X). Returning

to our

situation,

we have a map of sets cp: T ~

Ao(X), given by t - (-1)d-1C1(Lt)d.

LEMMA

(2.1):

Under the above

conditions, ~ :

T ~

A0(X) is a regular

map

in

the following

sense: there exist

a finite

open

cover {Ui} of T, positive integers

mi, n,

and morphisms 1: : Ui ~ Smi(U)

X

Sni(U) giving

commuta-

tive

diagrams

(where 03B3mi,ni

are the natural

maps).

* We could have used the c-closedness of W-equivalence to directly prove the equality of the 2 maps; but the proof of c-closedness is more subtle.

(11)

PROOF: Since T is

quasi-compact,

we need

only

prove the

following:

for each t E T there exists a

neighbourhood Ut of t, positive integers

m,

n and a

morphism ft : Ut ~ sm(u)

X

Sn(U)

such that

cr 1 Ut

=

Ym, n

°

ft.

Let .Yé’E Pic X be

ample.

Then

p*1H~ Pic(X T)

is

relatively

am-

ple

for

p2 : X T ~ T.

Hence for

large n,

we may assume that Hn and

Lt ~ Hn

are very

ample, Hi(X, Hn) = 0

for

i &#x3E; 0,

and

(Rip2)*(L~p*1Hn)

= 0 for i &#x3E; 0.

By

the base

change

theorem

(see [Ml],

pg.

46)

if

k(t)

is the residue field of

(2t,T

then

where

and j :

X X

Spec Ot,T ~ X

T is the natural map. Fix d effective di- visors

H1,..., Hd

in the linear

system | Hn |

on X such that

Hl

n ...

~ Hd

is a finite set of

points

contained in

U;

in

particular

all

the

partial

intersections are proper i.e. dim

Hil

~ ...

~ Hik = d - k

for

any k

distinct indices

i1,..., ik. Next,

choose d divisors

D1, ... , Dd

in

1 | Lt~Hn |

on X such that all the intersections

Di1 ~ ... ~ Dik n Hil

~ ...

Hjl

are proper, and all zero dimensional intersection

cycles

are

supported

on U

(we

can choose the

Di inductively, using

the remark that if

X ~ PN, Y1,..., Ym ~ X

subvarieties: then the

general hyperplane

section of X intersects all the

Y properly).

Now

Ox(Di - Hi) ~ Lt,

so

the intersection class

(-1)d-1C1(Lt)d is just

the class associated to the

zero

cycle

the

cycle

is well defined and

supported

on U

by

choice. Let

sl, ... , sd E

H0(X, Lt~Hn)

be sections

corresponding

to the divisors

D.

From

the consequence of the base

change

theorem noted

above,

we can

find a

neighbourhood

V of t ~ T and sections

Si E H0(X V, L~p*1Hn | X|V)

such

that si

restricts

to sl on X {t}.

If

b,

is the divisor of

st,

and

Iii

=

Hi X V,

then all the intersections

DI1 ~ ... ~Dik ~ Hj1 ~ ... ~Hjl

are proper, at least on some smaller open set of the form X

W, t ~

W c V

(since

the dimension of the fibers of a

surjective

proper

morphism

of varieties is an upper semi-continuous function on the base

[Hl]

II Ex.

3.22). By shrinking

W further we can

ensure that for k + 1

= d,

all intersections

If

r = r(il, ... , ik; j1,..., jl) = (Di1 ~ ... ~ Dik ~ Hj1 ~

...

~ Jl)

n

(X W),

where k + 1 =

d,

then 0393 ~ W is flat and proper of relative di-

(12)

mension 0

(flatness

is

immediate

from the Koszul resolution for the ideal sheaf of r c X X

W;

note that r c U X W where U is

smooth).

Since W

is

normal,

if 03A8 : r - W has

degree

q, then we have an induced mor-

phism W ~ Sq(U) given by w ~ [03A8-1(w)]

as an effective

0-cycle

of

degree q

on U. Give two

morphisms

g : W -

Sq(U),

h : W -

sr(u),

we

have a

morphism g + h : W ~ Sq+r(U)

obtained

by

addition of

0-cycles, (g + h)(w) = g(w) + h(w).

Hence for suitable

positive integers

m, n we

have a

morphism f W :

W -

sm(u)

X

Sn(U)

such that

as

0-cycles,

where the zero

cycle

A - B with

A,

B effective of

degrees

m,

n

corresponds

to

(A, B) ~ Sm(U)

X

Sn(U). Thus Ut = W, h = fW

have

the

required properties.

COROLLARY

(2.2):

Let X be a normal

projective variety

over an

algebrai- cally

closed

field

k. Then there exists a countable

family of

varieties

Wi, pairs of positive integers

mi, ni and

morphisms

gi :

wi ~ sm,(u)

X

,Snt(U)

such

tha~ if fi = 03B3mi,ni ° gl, fl : Wi ~ A0(X),

then

( i ) Ufi(Wi) ~ A0(X)

contains the rational

equivalence

classes

of

all

0-cycles of

the

form (-1)d-1C1(L)d,

cPE Pic

X,

d = dim X.

( ii ) dim W = q is independent of

i.

PROOF:

By

a result of

Chevalley [C],

there is an abelian

variety A

and

E9OE

Pic(X A)

such that

(A, Y)

is a Picard

family

i.e. if

Pic0X ~

Pic X

is the

subgroup

of bundles

algebraically equivalent

to

0,

then the map

A -

Pic X

given by a ~ La

= L

|X {a} gives

an

isomorphism

of groups

A - Pic°X; further,

any

family

of line bundles on X

parametrized by

a

normal

variety

T is induced

by

a

morphism

T -A. The Neron-Severi group

NS( X)

= Pic

X/Pic°X

is

finitely generated (see

Kleiman

[K],

for

example).

Thus if

f OE

Pic X

give

a

(countable)

set of coset

representa-

tives for Pic

X/PicoX,

then

p*1Lj~L~ Pic(X A) give

a countable set

of families of line bundles

on X,

each with

parameter

space

A,

such

that every line bundle on X occurs

exactly

once in a

unique family.

Corresponding

to each such

family,

we can associate a finite cover

{Wij}

of A and

morphisms gij : Wij ~ SmiJ(U) XniJ(U)

such that

VmiJniJ

o

gij(w)

is a zero

cycle representing (-1)d-1C1(Lw)d,

where

Lw = p*1Li~L|X {w}.

We can reindex the

{Wij}i,j by

the

positive

integers, giving

a set

{Wi}

of varieties with all the claimed

properties,

with

dim W

= q = dim

Pic°X.

§3.

Proof of the Theorem In fact we can prove a

slightly stronger

result.

(13)

THEOREM: Let

X/k be a projective variety of

dimension

d

2 over a universal domain with isolated normal

singularities.

Let V =

Spec

A be an

affine

open subset

of

X

containing

the

singular

locus.

Suppose A0(V) ~

0

( equivalently FoKo(V)

=1=

0).

Then there exist

(uncountably many) projec-

tive A-modules P

of

rank d such that P cannot be written as

Q

E9 L with

LEPicA.

We note that this

implies

the theorem stated in the

introduction,

since the

hypothesis

of that theorem

imply

that V =

Spec

A can be

compacti-

fied to a

projective variety X

such that

Xsing

c V

(this

uses resolution of

singularities).

Thus

F0K0(V) ~

0

implies

the existence of

indecomposa-

ble

projectives. Conversely,

if

FoKo(V)

= 0

and d 3,

then every

projec-

tive module of rank d has a trivial direct summand i.e. a unimodular element. This is obvious for d = 1 and follows from the

Murthy-Swan

cancellation theorem if d = 2

[MS].

If

d = 3,

this is the main result of

[MK],

if A is smooth over

k;

it has been

generalized

to

arbitrary

normal

A in

[L2].

We note that the conditions

F0K0(V) ~

0 and

A0(V) ~

0 are

equivalent -

indeed we have

surjections 03A8 : A0(V) ~ F0K0(V)

and

Cd : F0K0(V) ~ A0(V)

such that ’IF 0

Cd

and

Cd 0 ’IF

both

equal

multi-

plication by (-1)d-1(d - 1)!;

but both groups are divisible.

PROOF OF THE THEOREM: In the notation of

§1,

set

AW0(X) = A0(V).

Since X is

projective

and V ~ X is an affine open

subset,

X - V

generates

Alb X

(if

Y c X is a

general

2-dimensional linear space

section,

then Alb Y - Alb X

[La],

and

Y - (V ~ Y) is the complement

of an affine open set in Y which thus

supports

an

ample

divisor on Y

[H2]).

Since

AW0(X) ~ 0, by

Lemma

(1.5)

we have

dW

2 i.e. for m + n

n0,

In

particular,

for

large n,

we

have dn (= dn,o)

&#x3E; q = dim

Pic°X,

the

number

given by Corollary (2.2).

We claim that there exist elements of

~n(Sn(U)) ~ AW0(X) = A0(V)

which are not of the form

(- 1)1-ICI(y)d

for any E9OE Pic X.

Indeed,

we have

only

to

verify

that

U h - fi(Wi)

does not contain

~n(Sn(U)),

where h

= A0(X) ~ AW0(X) is the

natural

map. The set

is a

c-locally

closed set. We claim that

(14)

which will prove our earlier claim. Assume instead that

Up2(0393i,n) =

Sn(U).

Then some irreducible

component W

of the c-closure of some

rl, n

dominates

Sn(U)

under the

projection 0393i,n

z

Sn(U). By

definition of

d n ,

there exists a c-closed subset

F c Sn(U)

such that

if x ~ Sn(U) - F,

then every irreducible

component

of the c-closure of

~-1n(~n(x))

has

dimension

nd - d n .

Thus if

y ~ W

such that

image ( y ) E Sn(U) - F,

then the fiber of

W ~ Wi through y

has

dimension

nd -

d n .

Hence

dim Wi

dim W -

(nd - dn)

dim

Sn(U) - nd + dn =d n

&#x3E; q = dim

Wi,

a contradiction.

Thus,

the theorem will be

proved

once we show that there exists a

projective

A-module of rank d = dim A with a

prescribed top

Chern class in

A0(V), V = Spec A,

and

vanishing

lower Chern

classes,

with

respect

to a suitable

theory

of the Chow

ring

and Chern classes. For

non-singular V,

there is a standard

theory (see [F]).

We

briefly

review the

properties

of Levine’s

theory [L1]

which are needed.

First, CH1(V)

=

Pic V and

CHd(V) = A0(V) (the

second

equality

uses the

normality

of

V).

The first Chern class map

Ci

satisfies

C1(Ox(D))

=

[D] ~ CH1(X)

as usual. If

P, Q

are

projective modules, C(P), C(Q) ~ CH’(V)

the

total Chern

classes,

then

C( P

ei

Q )

=

C(P), C(Q).

The ’Riemann-Roch theorem without denominators’ is valid in the

following

sense: there is a

filtration FiK0(V)

and

maps 03A8i : CHi(F) ~ FiK0(V)/Fi+1K0(V)

such

that if

Cl : FiK0(V)/Fi+1K0(V) ~ CHi(F)

is induced

by

the i th Chern class

Ci, then ’¥i 0 Cl

and

Ci ° ’¥i

are both

multiplication by (-1)i-1(i -

1) !.

For a

subvariety

Z ~ V with Zn

Ving

=

0,

codim Z =

i,

there is a

class

[Z] ~ CHi(V)

such that

Ci(OZ) = (-1)i-1(i-1)![Z]

and

Cj(OZ)

= 0

for j i. Lastly,

if

D1,...,Dd ~ V

are effective Cartier divisors

meeting properly

with

(~Di)

n

V5;ing = ff,

then

[D1] · [D2]...[Dd] = [~Di ]

E

CHd(V)

=

A0(V).

These

properties

suffice for our purpose.

We now show that

given

any

03B4 ~ A0(V),

there exists a

projective

module P with rank P =

d, C1(P) = ··· = Cd-1(P)

= 0 and

Cd(P)

= 8.

If I c A is the ideal of an effective

(reduced) 0-cycle supported

on

V - Vsing,

then there is a resolution

where

F

are free and P’ is

projective.

Then

Ci(P’)

= 0 for i

d,

and

Cd(P’) = (-1)d-1Cd(A/I) = (d - 1)![A/I].

But any

0-cycle

is ration-

ally equivalent

to a

cycle

of the form

(d - 1)![A/I ]

for an effective

reduced

cycle [A/I],

since any zero

cycle

on V is

supported

on a smooth

affine curve C with C n

vsing

=

Ø, and

a similar claim is true for Pic C.

Thus we can find a

projective

module P with

C;(P’) = 0

for

i d,

Cd ( P’ )

= 03B4. Now if rank P’ = n

d, by

the

stability

theorem of Bass

[Ba]

we can write

(15)

where P has rank d. Then

q(P)

= 0 for i

d, Cd ( P )

= 8.

Applying

this

construction to a

class 03B4 ~ ~n(Sn(U)) - ~ h ° fi(Wi)

as

above,

we see

that P is

indecomposable.

This proves the

theorem.

§4. Examples

in dimension 2

The

examples

of affine varieties of

dimension

2 with nonzero Chow group of

0-cycles

are

given by

infinite

dimensionality

theorems for Chow groups of

projective

varieties. The best result is in dimension 2. Let X be

a normal

projective

surface over a universal domain k. We say that

Ao(X)

is

infinite

dimensional if none of the maps

’ym,n

of

§1

are

surjective. Equivalently,

the number

d(X)

=

dR(X) given by Prop. (1.2)

is

positive.

Then in fact

d(X) = 2, by

Lemma

(1.3).

One can

improve

lemma

(1.4)

as follows:

d(X) = 0 ~ A0(X)’ ~

Alb X

(where A0(X)’

~

AO(X)

is the

subgroup generated by cycles

of

degree 0).

Here Alb X is

universal for rational maps from X to abelian

varieties;

in

particular

it is

a birational invariant. Thus Alb X = Alb Y for any resolution of

singu-

larities Y ~ X. In

fact,

one can show that

d(X) = 0 ~ the composite A0(X) ~ A0(Y)’ ~ Alb

Y = Alb X is an

isomorphism (see [Sl],

Theo-

rem

1).

But one

easily

sees that if U c X is an affine open set

containing

Xsing, then A0(X)’ ~

Alb

Y ~ A0(U)

= 0

(see

loc. cit.: the

point

is that

X - U

supports

an

ample divisor).

In

particular, A0(U) ~ 0

if either

A0(Y)’ ~

Alb Y

(i.e. A0(Y)

is infinite

dimensional)

or

A0(X) ~ A0(Y)

is not

injective.

For smooth surfaces in

arbitrary

characteristic

(generalising

Mumford

[M2]

for k

= C)

Bloch

[Bl]

has shown that

A0(F)

is infinite dimensional if

and he also

conjectures

the converse. If k =

C,

this is

equivalent

to the

condition

r( Y, 03A92Y) ~

0. If X is a normal

projective surface,

it is shown

in

[Sl] that A0(X) ~ A0(Y) is

not

injective

in the

following

cases. Let

E c Y be the reduced

exceptional divisor,

which we may assume

(blow

up Y

further)

has smooth

components

and normal

crossings.

Then

(i)

If k

= C,

and

Pic0Y ~ Pic0(nE)

is nor

surjective

for some n &#x3E;

0,

then

A0(X) ~ A0(Y) is

not

injective.

(ii)

if k is a universal domain of

arbitrary characteristic,

and

Pic0 Y ~ Pic0E

is not

surjective,

then

A0(X) - A0(Y)

is not

injective.

The condition

(i)

is

equivalent

to the condition that

H2(X, OX) ~ H2(Y, OY)

is not

injective, by

the

Leray spectral

sequence and the formal function theorem. Thus if k =

C, A°(X)

is infinite dimensional

provided H2(X, (9x) *

0

(we give

another

proof

of this without K-the-

ory in

[S2]).

We

conjecture

that the converse

holds,

and that the converse

to

(ii)

is also valid.

Assuming

the above

conjectures

we have a

fairly

clear

Références

Documents relatifs

“naturally defined subsets” of abelian varieties (see Definition 2.1), assuming they are proper subsets for very general abelian varieties of a given dimension g, are at most

From this case the general dimension formula for affine Deligne–Lusztig varieties for special maximal compact subgroups of split groups follows, as was shown in a recent paper by

In other words, the differential form γ n is uniquely defined by the vanishing condition at ∞ and the secondary data dγ n = ν n+1 (see Remark 5.2). Do such primitives for more

It will follow that X(fe)/B is finite for any smooth rational surface over a global field, and that all points are Brauer equivalent at good reduction places.. This explains

Our result is the classification of the G-vector bundles over X up to G-equivariant isomorphism.. This type of classification problems have been introduced, in a

In particular we give the proof of a statement of Deligne which generalizes the statement of theorems of Lefschetz type to finite morphisms into a projective space.. The proofs

Throughout this section X is a projective algebraic variety, on which an algebraic torus T acts linearly having only a finite number of invariant

component of the Chow variety d,r whose general member is irreducible and nondegenerate is achieved by a component parametrizing curves lying.. on a rational